Palantir Is Down: What I Learned From a First Mini Research

In September 2025 I wrote up my first stock and called Palantir a short. The case was simple: at 567 times earnings and 118 times sales, the price was buying a story, not a business.¹ The story has since come apart. Palantir trades near $132, down about a third from where the market crowned it and well off its $207 peak,² and the hype I pointed at has drained out of the multiple.

When I published, almost everyone disagreed. The large majority of analysts covering Palantir rated it a buy, with an average target near $180, roughly where it traded.³ I am not an analyst; I was a beginner writing my first stock pitch for fun, betting against what the professionals thought. The tape since has gone the skeptic's way.

Being early looks like being wrong

The call did not pay right away. Palantir climbed to a record close near $207 before it fell, about 21 percent above where I called it,² so anyone short at $171 would have been underwater for months. There is an old line, usually attributed to Keynes: the market can remain irrational longer than you can remain solvent. Fading hype works like that; the price runs past reason first, and being early looks exactly like being wrong for a long time before you can tell the two apart. I am still not sure I can, and that is part of the lesson.

The hype was the thesis

My case was never that Palantir was a bad company; it was that the price had detached from any business that could grow into it. Bulls will say the numbers proved me wrong, because revenue accelerated instead of slowing, up 85 percent in the first quarter with profit quadrupling.⁴ The growth is real. It is also the wrong kind: it came from leaning harder into government, defense, and immigration work, the most politically exposed revenue a software company can carry, not the broad commercial base that would justify the multiple. Palantir did not outgrow the hype; it cashed it in the fastest, most fragile way it could.

The clearest evidence is the multiple itself. At the call Palantir traded at 118 times sales; today it trades at 59, roughly half, even after revenue doubled, and still about seven times its peers.⁵ That compression is what I was short, and it may not be finished, though I have learned to distrust my own confidence about how much further a thing can fall. Some of it was the whole software sector derating on AI fears⁶ and a well-publicized short from Michael Burry,⁷ and the stock has since bounced on an Nvidia headline.¹⁰ But the lasting damage is coming from something the deck only gestured at: what Palantir chose to become.

The money it chose to chase

Palantir's revenue is concentrated where it is most exposed: about 53 percent government and 79 percent United States.¹¹ That is not a moat; it is a bet that one set of political customers keeps buying. To keep them buying, the company took work most software vendors will not touch.

Its edge is a willingness to take on immigration enforcement, intelligence, and battlefield AI, and management does not hedge about it. Alex Karp answered a protester who accused the company of killing Palestinians with its AI by saying "mostly terrorists," and has told critics of ICE they "should be out there protesting for more Palantir."¹⁷ That posture wins contracts and draws opposition an ordinary software vendor never has to manage.

The opposition is concrete. Palantir built an ICE system called ImmigrationOS to help identify deportation targets,¹³ and a Palantir tool was reported to mine Medicaid records for leads.¹⁴ Former employees have objected, signing an open letter against its Trump-administration ties.¹⁵ Abroad, Amnesty International and a UN special rapporteur have tied its technology to Israel's operations in Gaza, allegations the company rejects.¹⁶

This is turning into lost revenue. France's DGSI is moving to a domestic vendor, Germany's domestic-intelligence agency chose a French supplier over Palantir, Switzerland has rejected its tools on data-sovereignty grounds, London's mayor blocked a £50 million Met Police contract, and Parliament has pushed to exit the £330 million NHS deal.^{8, 9, 18, 19, 20} Several landed in the June window that finally pushed the stock below my entry.

A Financial Times investigation put it plainly: the pushback to Palantir's politics may threaten the core of its $330 billion business.¹⁸ It found more than sixty senior engineers gone to AI labs in a year, big investors trimming stakes over the company's politics, and a political clock ticking: a Democratic House could subpoena its executives, and a Democratic White House in 2028 could put more than $10 billion of defense contracts in play. Palantir disputes parts of the account.

The quieter drag is dilution. Share count has climbed from about 1.9 billion in 2021 to 2.6 billion, an increase of about a third, all from stock-based pay rather than a capital raise, so the same business keeps getting split across more shares.¹²

The strongest case against me

The honest way to hold a view is to argue the other side as hard as you can. Growth reaccelerated to 85 percent, which is the best argument against me, but it is government-led and politically concentrated, not the durable base that justifies the price. The forward-deployed model is a real moat, but an increasingly political one, and politics cuts both ways. The stock bounced, but on a headline, the same kind of story that inflated it in the first place. Karp's bombast once drew talent and customers; now the engineers and the capital are leaving.

Where this leaves it

The thesis was that Palantir's price was hype, and the hype has cracked. The harder question was always what sat underneath, and the answer is a business that runs on the most politically fragile revenue it can find, a worse foundation at $132 than the market believed at $171. I would think differently if the commercial base could stand without the government engine, if the political pressure eased instead of building, and if the multiple fell to something a normal software company can defend. None of that has happened yet.

I am keeping the original deck up, unedited, next to this note.¹ The point was never to be proven right; it was to make a real call and follow it honestly. What I take from it is smaller than a win: a company can grow and weaken at the same time, and being early is hard to separate from being wrong until long after the fact. For a first attempt at real research, that lesson is worth more than the call. I do this to learn, not to advise.